Method and system for predicting stomach cancer immunotherapy and chemotherapy combined curative effect based on CT image, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of biological medicines, and particularly discloses a method and a system for predicting a gastric cancer immunotherapy and chemotherapy combined curative effect based on a CT image, electronic equipment and a storage medium. The method for predicting the stomach cancer immunotherapy and chemotherapy combined curative effect based on the CT image comprises the following steps: S1, processing a medical image of a stomach tumor area of a stomach cancer patient, and obtaining an image of an area of interest of a focus of the medical image of the tumor area; s2, extracting radiomics features in the image of the region of interest; s3, inputting the values of the extracted image omics features into an image score prediction model, and calculating to obtain an image score of the region-of-interest image; and S4, comparing the image score with a preset threshold value, and predicting the immunotherapy and chemotherapy combined curative effect of the gastric cancer patient according to the size relationship between the image score and the preset threshold value. The method disclosed by the invention can be used for effectively predicting the curative effect of immunotherapy and chemotherapy combined with the gastric cancer patient, and has relatively high accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the fields of biomedical technology and medical image processing, and particularly discloses a method and system for predicting the curative effect of immunotherapy combined with chemotherapy for gastric cancer based on CT images, an electronic device and a storage medium. BACKGROUND

[0002] Gastric cancer is the first digestive tract malignant tumor in terms of incidence in China, and has a complex pathogenesis and high malignancy, seriously threatening public health safety. For early gastric cancer, the treatment method mainly based on surgical resection can achieve good curative effect. However, due to the lack of public medical health knowledge, the imperfect early screening system and the non-obvious symptoms of early disease, about 60%-70% of gastric cancer patients are in the progressive or advanced stage at the time of diagnosis, which may miss the opportunity of radical surgical resection, resulting in poor prognosis. For such gastric cancer patients, the current clinical treatment means is very limited, and the chemotherapy scheme mainly based on the combination of fluorouracil drugs and platinum drugs can prolong the survival period to a certain extent, but the median overall survival is less than 1 year, and the curative effect is still not ideal. In addition, the toxic and side effects of chemotherapy drugs seriously affect the quality of life of patients and reduce the tolerance and compliance of patients. Therefore, it is urgent to develop new treatment strategies to improve the prognosis of patients with unresectable gastric cancer.

[0003] In recent years, the emergence of PD-1 / PD-L1 and other ICIs has brought new hope for the treatment of malignant tumors such as advanced gastric cancer. PD-1 / PD-L1 inhibitors can reactivate the patient's anti-tumor immune response by blocking the inhibitory signal pathway of the immune system of tumor cells, and have achieved remarkable breakthroughs in the treatment of various solid tumors. Clinical studies have confirmed that compared with traditional chemotherapy, immunotherapy (PD-1 / PD-L1 inhibitor treatment) can significantly prolong the overall survival of some patients with advanced gastric cancer. In the KEYNOTE-062 clinical trial, for patients with advanced gastric cancer with positive PD-L1 expression, the median overall survival of the group using pembrolizumab combined with chemotherapy was 17.4 months, while the median overall survival of the simple chemotherapy group was only 10.8 months. However, the response rate of immunotherapy has significant individual differences, and how to accurately identify the benefit population and develop individualized treatment plans is a key problem to be solved in the field of tumor immunotherapy. At present, the PD-L1 expression level, tumor mutation load and microsatellite instability status and other biomarkers have shown potential in predicting the curative effect of immunotherapy (PD-1 / PD-L1 inhibitor treatment) to a certain extent, but the existing detection methods have limitations such as invasiveness, influence of tumor heterogeneity, non-uniform detection means, long cycle, high price, etc., which are difficult to realize rapid, convenient and low-cost promotion and popularization in clinical practice.

[0004] With its non-invasiveness and simplicity, imaging biomarkers have attracted much attention in the diagnosis, treatment and prognosis of gastric cancer, and provide hope to overcome the limitations of traditional markers. CT, as a mature imaging examination method, can comprehensively evaluate tumor information and is the preferred method for the diagnosis, efficacy monitoring and follow-up of gastric cancer. In recent years, CT-based imaging genomics technology has developed rapidly. By combining advanced image processing and machine learning techniques, quantitative evaluation of tumors and their microenvironments can be achieved, and the integration of patient clinical, histological characteristics, genetic information and other multi-dimensional characteristics can provide more assistance for individualized efficacy evaluation and prognosis prediction. However, the current CT imaging genomics research on the efficacy prediction of gastric cancer immunotherapy (PD-1 / PD-L1 inhibitor treatment) combined with chemotherapy is still limited, which restricts the application of CT imaging genomics in individualized immunotherapy of gastric cancer. SUMMARY

[0005] In view of the problems and deficiencies in the prior art, the purpose of the present application is to provide a method and system for predicting the efficacy of gastric cancer immunotherapy combined with chemotherapy based on CT imaging, an electronic device and a storage medium.

[0006] To achieve the purpose of the application, the technical solutions adopted by the present application are as follows:

[0007] The present application provides a method for predicting the efficacy of gastric cancer immunotherapy combined with chemotherapy based on CT imaging, comprising the following steps:

[0008] S1, processing the medical image of the tumor region of the stomach of the gastric cancer patient to obtain the region of interest image of the lesion of the tumor region medical image; wherein the medical image is a CT image;

[0009] S2, extracting the imaging genomics features in the region of interest image;

[0010] S3, inputting the values of the imaging genomics features extracted in step S2 into an image score prediction model to calculate the image score of the region of interest image;

[0011] S4, comparing the image score obtained in step S3 with a preset threshold value, and predicting the efficacy of the immunotherapy combined with chemotherapy of the gastric cancer patient according to the size relationship between the image score and the preset threshold value.

[0012] According to the above method, preferably, in step S2, the imaging genomics features are as follows:

[0013] Original shape Flatness,

[0014] Wavelet LLH glcm Imc1,

[0015] Wavelet LLH ngtdm Busyness,

[0016] Wavelet LHL glcm Autocorrelation,

[0017] Wavelet LHH firstorder Kurtosis,

[0018] Wavelet HLL glcm MaximumProbability,

[0019] Wavelet HLL glszm GrayLevelNonUniformity,

[0020] Wavelet HLL ngtdm Contrast,

[0021] Wavelet HLH firstorder Kurtosis,

[0022] Wavelet HLH firstorder Median,

[0023] Wavelet HHL glcm ClusterShade,

[0024] Wavelet HHL glcm InverseVariance,

[0025] Wavelet HHL glszm GrayLevelVariance,

[0026] Wavelet LLL firstorder Maximum,

[0027] Wavelet LLL glszm GrayLevelNonUniformityNormalized.

[0028] According to the method described above, preferably, the image score prediction model is used to calculate the calculation formula of the image score:

[0029] Radscore = 0.049124367 - 0.108127438 x Original shape Flatness - 0.00194955 x Wavelet LLH glcm Imcl - 0.011553971 x Wavelet LLH ngtdm Busyness - 0.023343715 x Wavelet LHL glcm Autocorrelation + 0.070103348 x Wavelet LHH firstorder Kurtosis + 0.032363627 x Wavelet HLL glcm MaximumProbability - 0.271672778 x Wavelet HLLglszm GrayLevelNonUniformity - 0.237148842 x Wavelet HLL ngtdm Contrast + 0.205449843 x Wavelet HLH firstorder Kurtosis + 0.262356661 x Wavelet HLHfirstorder Median - 0.073857162 x Wavelet HHL glcm ClusterShade + 0.024585599 x Wavelet HHL glcm InverseVariance - 0.022673688 x Wavelet HHL glszm GrayLevelVariance - 0.062524036 x Wavelet LLL firstorder Maximum + 0.187431418 x Wavelet LLL glszm GrayLevelNonUniformityNormalized.

[0030] According to the method, preferably, in step S4, when the image score is greater than a preset threshold, the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be effective (the effective means that the efficacy evaluation of the immunotherapy combined with chemotherapy for the gastric cancer patient is immune complete remission or immune partial remission); when the image score is less than or equal to the preset threshold, the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be ineffective (the ineffective means that the efficacy evaluation of the immunotherapy combined with chemotherapy for the gastric cancer patient is immune stable disease or immune confirmed disease progression). More preferably, the preset threshold is 0.54. Further, when the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be effective, the gastric cancer patient is recommended to use the immunotherapy combined with chemotherapy; and when the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be ineffective, the gastric cancer patient is not recommended to use the immunotherapy combined with chemotherapy.

[0031] According to the method described above, preferably, the gastric cancer is unresectable gastric cancer (i.e. non-surgically resectable gastric cancer).

[0032] According to the method described above, preferably, the immunotherapy is treatment with a PD-1 inhibitor or / and a PD-L1 inhibitor; more preferably, the immunotherapy is treatment with a PD-1 inhibitor, including nivolumab, pembrolizumab, Tislelizumab, atezolizumab, durvalumab, avelumab, nivolumab, pembrolizumab.

[0033] According to the method described above, preferably, the CT image is a venous phase CT image.

[0034] According to the method described above, preferably, the region of interest image is a two-dimensional region of interest image.

[0035] According to the method described above, preferably, the chemotherapy is treatment with a chemotherapeutic drug, including oxaliplatin, capecitabine, oxaliplatin, calcium folinate, levofolinate, 5-fluorouracil. More preferably, the chemotherapy administration includes two schemes: 1) XELOX scheme: oxaliplatin and capecitabine are administered in combination (oxaliplatin is administered on day 1 of each treatment cycle, and capecitabine is administered continuously from day 1 to day 14 of each treatment cycle); 2) FOLFOX scheme: oxaliplatin is administered in combination with calcium folinate (or levofolinate) and 5-fluorouracil, wherein oxaliplatin is administered on day 1 of each treatment cycle, calcium folinate (or levofolinate) is administered on day 1 of each treatment cycle, and 5-fluorouracil is administered within the first 46 hours after the start of each treatment cycle (preferably, 5-fluorouracil is administered intravenously at a dose of 400 mg / m 2 ) on day 1 of each treatment cycle, followed by a continuous intravenous infusion of 2400-3600 mg / m 2 ) for 46 hours.

[0036] The second aspect of the present application provides a prediction system for predicting the curative effect of immunotherapy combined with chemotherapy for gastric cancer based on CT images, which comprises an image input module, an image processing module, a feature extraction module, a score prediction module, and an image classification module; wherein the image input module is used for inputting a medical image of a tumor region in the stomach of a gastric cancer patient, the image processing module is used for processing the input medical image of the tumor region in the stomach to obtain a region of interest image of the lesion in the tumor region medical image; the feature extraction module is used for extracting imaging features in the region of interest image; the score prediction module is provided with a score prediction model, which is used for calculating the image score of the region of interest image according to the values of the imaging features extracted by the feature extraction module; and the prediction and result output module is used for qualitatively analyzing the image score, predicting the curative effect of immunotherapy combined with chemotherapy for the gastric cancer patient, and outputting the prediction result.

[0037] According to the prediction system described above, preferably, the imaging features are as follows:

[0038] Original shape Flatness,

[0039] Wavelet LLH glcm Imc1,

[0040] Wavelet LLH ngtdm Busyness,

[0041] Wavelet LHL glcm Autocorrelation,

[0042] Wavelet LHH firstorder Kurtosis,

[0043] Wavelet HLL glcm MaximumProbability,

[0044] Wavelet HLL glszm GrayLevelNonUniformity,

[0045] Wavelet HLLngtdm Contrast,

[0046] Wavelet HLH firstorder Kurtosis,

[0047] Wavelet HLH firstorder Median,

[0048] Wavelet HHL glcm ClusterShade,

[0049] Wavelet HHL glcm InverseVariance,

[0050] Wavelet HHL glszm GrayLevelVariance,

[0051] Wavelet LLL firstorder Maximum,

[0052] Wavelet LLL glszm GrayLevelNonUniformityNormalized.

[0053] According to the prediction system described above, preferably, in step S3, the image score prediction model is used to calculate the calculation formula of the image score:

[0054] Radscore = 0.049124367 - 0.108127438 x Original shape Flatness - 0.00194955 x Wavelet LLH glcm Imcl - 0.011553971 x Wavelet LLH ngtdm Busyness - 0.023343715 x Wavelet LHL glcm Auto correlation + 0.070103348 x Wavelet LHH firstorder Kurtosis + 0.032363627 x Wavelet HLL glcm Maximum Probability - 0.271672778 x Wavelet HLL glszm Gray Level Non Uniformity - 0.237148842 x Wavelet HLL ngtdm Contrast + 0.205449843 x Wavelet HLH firstorder Kurtosis + 0.262356661 x Wavelet HLH firstorder Median - 0.073857162 x Wavelet HHL glcm Cluster Shade + 0.024585599 x Wavelet HHL glcm Inverse Variance - 0.022673688 x Wavelet HHL glszm Gray Level Variance - 0.062524036 x Wavelet LLL firstorder Maximum + 0.187431418 x Wavelet LLL glszm Gray Level Non Uniformity Normalized.

[0055] According to the prediction system, preferably, in the step S4, the prediction and result output module is configured to qualitatively analyze the image score, predict the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient, and output the prediction result, and the specific operation is that: comparing the image score with a preset threshold value, predicting the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient according to the size relationship between the image score and the preset threshold value; when the image score is greater than the preset threshold value, predicting that the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is effective (the effective means that the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is evaluated as immune complete remission or immune partial remission); and when the image score is less than or equal to the preset threshold value, predicting that the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is ineffective (the ineffective means that the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is evaluated as immune stable disease or immune confirmed disease progression). More preferably, the preset threshold value is 0.54. Further, when the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is effective, it is recommended to use the immunotherapy combined with the chemotherapy for the gastric cancer patient; and when the efficacy of the immunotherapy combined with the chemotherapy for the gastric cancer patient is ineffective, it is not recommended to use the immunotherapy combined with the chemotherapy.

[0056] According to the prediction system, preferably, the gastric cancer is unresectable gastric cancer (i.e., non-surgical resection gastric cancer).

[0057] According to the method, preferably, the immunotherapy is treatment using a PD-1 inhibitor or / and a PD-L1 inhibitor; more preferably, the immunotherapy is treatment using a PD-1 inhibitor, and the PD-1 inhibitor includes nivolumab, pembrolizumab, Tislelizumab, atezolizumab, durvalumab, avelumab, nivolumab, pembrolizumab.

[0058] According to the method, preferably, the CT image is a CT image in the venous phase.

[0059] According to the method, preferably, the region of interest image is a two-dimensional region of interest image.

[0060] According to the method, preferably, the chemotherapy is treatment with a chemotherapy drug, and the chemotherapy drug includes oxaliplatin, capecitabine, oxaliplatin, calcium folinate or leucovorin calcium, 5-fluorouracil. 2 Then, intravenous infusion of 2400-3600 mg / m 2 ) is performed for 46 hours.

[0061] According to the system, preferably, the image processing module processes the gastric tumor area medical image by using open source software 3D Slicer (version 4.10.0) to segment a region of interest (ROI) image of the lesion in the gastric tumor area medical image.

[0062] The fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is loaded by a processor to execute the method for predicting the curative effect of immunotherapy combined with chemotherapy on gastric cancer based on CT images.

[0063] Compared with the prior art, the present application has the following beneficial effects:

[0064] (1) The present application obtains the ROI region image of the tumor lesion by processing the CT image of the tumor lesion of the gastric cancer patient, extracts the radiomics features of the ROI region image, and then predicts the efficacy of the immunotherapy combined with chemotherapy of the gastric cancer patient according to the extracted radiomics features. Therefore, the method of the present application can effectively predict the efficacy of the immunotherapy combined with chemotherapy of the gastric cancer patient, and the AUC value of the predicted efficacy in the validation set is 0.816 (95% CI: 0.728-0.904), the accuracy is 80.4%, the sensitivity is 70.0%, and the specificity is 91.5%. It can be seen that the prediction method of the present application has high accuracy, not only provides a new means for predicting the efficacy of the current clinical immunotherapy combined with chemotherapy for unresectable gastric cancer, but also provides a reference for formulating a precise immunotherapy scheme for unresectable gastric cancer, which is beneficial to the determination of a more precise treatment scheme for gastric cancer by the clinician. Moreover, as an image-based diagnostic technique, the present application has the advantages of non-invasiveness, intuitiveness, easy operation, high detection efficiency and high prediction accuracy, can reduce the pain and economic burden caused by invasive diagnosis, and also makes up for the shortcomings of the current conventional diagnostic method.

[0065] (2) The radiomics features extracted from the ROI region image are obtained from the tumor target lesion and are not interfered by individual factors such as age and gender, and have universality and generality. Therefore, the method of the present application is universal and has stable evaluation effect in different environments. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 It is a schematic diagram of ROI delineation. Wherein, A is a venous phase original image; B is a venous phase tumor delineation region of interest (green) image;

[0067] Figure 2 It is the image score (Radscore) score distribution of the gastric tumor region CT image of the gastric cancer patient in different data sets and the comparison of the image score between the effective group and the ineffective group. A and B are the training set; C and D are the internal verification set; E and F are the external verification set. In the figure, the radiomics score represents the image score (Radscore);

[0068] Figure 3 It is the ROC curve of the training set;

[0069] Figure 4 It is the ROC curve of the internal verification set (A) and the external verification set (B);

[0070] Figure 5 It is the calibration curve of the radiomics model and the Logistic model for predicting the efficacy of the immunotherapy combined with chemotherapy for gastric cancer. Wherein A is the training set; B is the internal verification set; C is the external verification set;

[0071] Figure 6 The decision curve of the imageomics model and the Logistic model; wherein A is a training set; B is an internal validation set; and C is an external validation set. DETAILED DESCRIPTION

[0072] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be further clearly and completely described below with reference to the embodiments of the present application. It should be noted that the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0073] The statistical methods used in the following embodiments of the present application are as follows: GraphPad Prism software (version: 10.1.2) and R software (version: 4.2.2) are used for statistical analysis, Shaprio-Wilk test is used to evaluate whether the quantitative data conforms to the normal distribution, the variables conforming to the normal distribution are represented by mean ± standard deviation ( ), the variables not conforming to the normal distribution are represented by median (quartile), the independent sample t test or Mann-Whitney U test is used for comparison between the treatment effective and ineffective groups. The frequency of the classification variable is represented, and the Pearson chi-square test, continuity correction test, or Fisher's exact probability method is used for comparison between groups. Bilateral P<0.05 is statistically significant. The main toolkits or functions of the R software are as follows: "icc" is used for calculation of the intra-class correlation coefficient ICC, "glmnet" package is used for logistic regression (including LASSO regression) analysis, "pROC" package is used for ROC analysis, "rmda" package is used for decision curve analysis, "calibration" function in "rms" package is used for calibration curve analysis, and "PredictABEL" package is used for NRI and IDI analysis.

[0074] Embodiment 1:

[0075] A method for predicting the efficacy of immunotherapy combined with chemotherapy for gastric cancer based on CT images, the specific steps are:

[0076] S1, acquiring a venous phase CT image of a tumor region in the stomach of a gastric cancer (preferably unresectable gastric cancer) patient, processing the venous phase CT image of the tumor region in the stomach of the gastric cancer patient, and acquiring a region of interest image of the lesion of the tumor region in the venous phase CT image;

[0077] S2, extracting the radiomics features in the two-dimensional region of interest image by using the Pyradiomics plug-in in the 3D Slicer software, to obtain the extracted radiomics features and the values thereof;

[0078] S3, inputting the values of the radiomics features extracted in step S2 into an image score prediction model, to calculate the image score of the two-dimensional region of interest image;

[0079] S4, comparing the image score obtained in step S3 with a preset threshold, and predicting the efficacy of immunotherapy combined with chemotherapy for the gastric cancer patient according to the size relationship between the image score and the preset threshold. The immunotherapy is treatment with a PD-1 inhibitor, and the PD-1 inhibitor includes nivolumab, pembrolizumab, Tislelizumab, atezolizumab, durvalumab, avelumab, nivolumab, pembrolizumab. The chemotherapy is treatment with a chemotherapeutic drug, and the chemotherapeutic drug includes oxaliplatin, capecitabine, oxaliplatin, calcium folinate or levofolinate, 5-fluorouracil. More preferably, the chemotherapy includes two schemes: 1) XELOX scheme: oxaliplatin and capecitabine are administered in combination (oxaliplatin is administered on day 1 of each treatment cycle, and capecitabine is administered continuously from day 1 to day 14 of each treatment cycle); 2) FOLFOX scheme: oxaliplatin is administered in combination with calcium folinate (or levofolinate) and 5-fluorouracil, wherein oxaliplatin is administered on day 1 of each treatment cycle, calcium folinate (or levofolinate) is administered on day 1 of each treatment cycle, and 5-fluorouracil is administered within the first 46 hours after the start of each treatment cycle (preferably, 5-fluorouracil is administered intravenously at a dose of 400 mg / m 2 ) on day 1 of each treatment cycle, followed by intravenous infusion of 2400-3600 mg / m 2 ) for 46 hours.

[0080] In step S1, the venous phase CT image of the tumor region of the stomach of the gastric cancer patient is processed, and the specific operation of obtaining the region of interest image of the lesion of the tumor region venous phase CT image is as follows: the venous phase CT image of the tumor region of the stomach of the gastric cancer patient (preferably unresectable gastric cancer) is preprocessed, and then the 3D Slicer software 3D Slicer (version 5.0.2, http: / / www.slicer.org) is used by an image diagnostician with more than or equal to 6 years of gastrointestinal tumor reading experience to manually segment and delineate the region of interest on the tumor maximum cross section of the preprocessed tumor region venous phase CT image along the lesion edge, and obtain the tumor region of interest image (ROI image). The preprocessing method is as follows: ① voxel resampling to 1mm×1mm×1mm; ② gray value discretization, group interval is 25; ③ gray scale normalization.

[0081] In step S2, the image features are as follows:

[0082] Original shape Flatness、

[0083] Wavelet LLH glcm Imc1、

[0084] Wavelet LLH ngtdm Busyness、

[0085] Wavelet LHL glcmAutocorrelation、

[0086] Wavelet LHH firstorder Kurtosis、

[0087] Wavelet HLL glcm MaximumProbability、

[0088] Wavelet HLL glszm GrayLevelNonUniformity、

[0089] Wavelet HLL ngtdm Contrast、

[0090] Wavelet HLH firstorder Kurtosis、

[0091] Wavelet HLH firstorder Median、

[0092] Wavelet HHL glcm ClusterShade、

[0093] Wavelet HHL glcm InverseVariance,

[0094] Wavelet HHL glszm GrayLevelVariance,

[0095] Wavelet LLL firstorder Maximum,

[0096] Wavelet LLL glszm GrayLevelNonUniformityNormalized.

[0097] In step S3, the image score prediction model is used to calculate the calculation formula of the image score:

[0098] Radscore = 0.049124367 - 0.108127438 x Original shape Flatness - 0.00194955 x Wavelet LLH glcm Imc1 - 0.011553971 x Wavelet LLH ngtdm Busyness - 0.023343715 x Wavelet LHL glcm Auto correlation + 0.070103348 x Wavelet LHH firstorder Kurtosis + 0.032363627 x Wavelet HLL glcm Maximum Probability - 0.271672778 x Wavelet HLL glszm GrayLevelNonUniformity - 0.237148842 x Wavelet HLL ngtdm Contrast + 0.205449843 x Wavelet HLH firstorder Kurtosis + 0.262356661 x Wavelet HLH firstorder Median - 0.073857162 x Wavelet HHL glcm ClusterShade + 0.024585599 x Wavelet HHL glcm InverseVariance - 0.022673688 x Wavelet HHL glszm GrayLevelVariance - 0.062524036 x Wavelet LLL firstorder Maximum + 0.187431418 x Wavelet LLL glszm GrayLevelNonUniformityNormalized.

[0099] In step S4, the image score is compared with a preset threshold value, and the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted according to the size relationship between the image score and the preset threshold value (the preset threshold value is preferably 0.54); when the image score is greater than the preset threshold value, it is predicted that the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is effective (the effective means that the efficacy evaluation of the immunotherapy combined with chemotherapy for the gastric cancer patient is immune complete remission or immune partial remission); when the image score is less than or equal to the preset threshold value, it is predicted that the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is ineffective (the ineffective means that the efficacy evaluation of the immunotherapy combined with chemotherapy for the gastric cancer patient is immune stable disease or immune confirmed disease progression). Further, when the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be effective, it is recommended that the gastric cancer patient uses the immunotherapy combined with chemotherapy; when the efficacy of the immunotherapy combined with chemotherapy for the gastric cancer patient is predicted to be ineffective, it is not recommended to use the immunotherapy combined with chemotherapy.

[0100] Further, the image features in step S2 are obtained through screening, and the image score prediction model in step S3 is a linear score prediction model established according to the image features screened in step S2. The specific method steps of the image feature screening in step S2 and the image score prediction model construction in step S3 are as follows:

[0101] 1. Research subjects, collection of clinical data of research subjects, treatment regimens and efficacy evaluation

[0102] (1) Research subjects

[0103] This retrospective study has been approved by the hospital ethics committee (ethics number: 2021-KY-1070-002) and exempted from signing informed consent. Retrospective and continuous collection of gastric cancer patients admitted to our hospital (hospital 1) from May 2019 to February 2024 and hospital 2 from January 2021 to January 2023. The clinical and imaging data of the patients were collected through the case system and the picture archiving and communication system (PACS), and further screened according to the following standards.

[0104] The inclusion criteria for the research subjects are: 1) pathologically and histologically confirmed as gastric adenocarcinoma; 2) unable to undergo surgical radical resection after multidisciplinary expert consultation; 3) received PD-1 / PD-L1 inhibitor combined with chemotherapy treatment; 4) received CT plain scan and dual-phase enhanced scan (baseline CT) within 1 week before combined treatment.

[0105] Exclusion criteria of the study subjects were: 1) received other anti-tumor therapy before combination therapy; 2) combination therapy was not completed for 3 cycles; 3) lack of measurable lesions or irregular review, which could not evaluate the efficacy; 4) the best efficacy evaluation was immune unconfirmed progressive disease (iUPD); 5) poor CT image quality or poor gastric filling, which affected lesion evaluation and delineation; 6) combined with other primary malignant tumors or serious systemic diseases.

[0106] According to the above inclusion and exclusion criteria, 328 patients were finally included in hospital 1, of which 235 were males and 93 were females, with an average age of 62.03±12.02 years and a median age of 64 years (range 27-86 years); 40 patients were finally included in hospital 2, of which 33 were males and 7 were females, with an average age of 60.68±9.99 years and a median age of 58 years (range 34-79 years). The 328 patients included in hospital 1 were divided into training set (n=231) and internal validation set (n=97) by stratified random sampling method in the ratio of 7:3, and the 40 patients included in hospital 2 were used as external validation set.

[0107] (2) Collection of clinical data of the study subjects

[0108] The clinical data of the patients were recorded and sorted through the medical record system, mainly including: 1) general information: age, gender, treatment cycle; 2) laboratory indicators: hemoglobin, platelets, neutrophils, lymphocytes, monocytes, neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-monocyte ratio (LMR) before treatment; 3) tumor markers: Carcinoembryonic antigen (CEA), Carbohydrate antigen 19-9 (CA19-9) and Carbohydrate antigen 72-4 (CA72-4) before treatment.

[0109] The statistical results of the clinical data of the patients in the training set, internal validation set and external validation set are shown in Table 1.

[0110]

[0111] (3) Treatment regimen and efficacy evaluation

[0112] 1) Treatment regimen

[0113] Patients were all treated with PD-1 / PD-L1 inhibitor combined with chemotherapy as the first-line treatment regimen, including:

[0114] A, Pembrolizumab combined with XELOX regimen, every 21 days as a treatment cycle, the administration method and dose are as follows: ① Pembrolizumab injection (Merck Sharp & Dohme Co., USA), fixed dose 200 mg, intravenous infusion, administered on the first day of each treatment cycle; ② XELOX regimen contains the following two drugs: oxaliplatin: the dose is 130 mg / m2, intravenous infusion, administered on the first day of each treatment cycle; capecitabine: the dose is 1000 mg / m2, oral administration, twice a day, from the first day to the 14th day of each treatment cycle.

[0115] B, Nivolumab combined with FOLFOX regimen, every 14 days as a treatment cycle, the administration method and dose are as follows: ① Nivolumab injection (Bristol-Myers Squibb Co., USA): fixed dose 240 mg, intravenous infusion, administered on the first day of each treatment cycle; ② FOLFOX regimen contains the following three drugs: oxaliplatin: the dose is 85 mg / m 2 , intravenous infusion, administered on the first day of each treatment cycle; calcium folinate or levofolinate: calcium folinate dose is 400 mg / m 2 , or levofolinate dose is 200 mg / m 2 , intravenous infusion, administered on the first day of each treatment cycle; 5-fluorouracil: first 400 mg / m 2 intravenous injection, then continuous intravenous infusion at a dose of 2400-3600 mg / m2 for the next 46 hours.

[0116] C, Nivolumab combined with XELOX, every 21 days as a treatment cycle, the administration method and dose are as follows: ① Nivolumab injection (Bristol-Myers Squibb Co., USA): fixed dose 360 mg, intravenous infusion, administered on the first day of each treatment cycle; ② XELOX regimen is the same as above.

[0117] D, Sintilimab combined with XELOX: ① Sintilimab injection (Sintong Biopharmaceutical Co., Ltd.): the dose is determined according to the body weight of the patient, <60 kg: 3 mg / kg; ≥60 kg: fixed dose of 200 mg; intravenous infusion, administered on the first day of each treatment cycle; ② XELOX regimen is the same as above.

[0118] 2) Efficacy evaluation

[0119] The efficacy was evaluated by immune Response Evaluation Criteria In Solid Tumors (iRECIST), and the details were shown in Table 2. The efficacy criteria included immune complete response (iCR), immune partial response (iPR), immune stable disease (iSD) and immune confirmed progressive disease (iCPD). The patients with iCR and iPR were the treatment effective group, and the patients with iSD and iCPD were the treatment ineffective group.

[0120]

[0121] The cycle range of the 231 samples in the training set treated by PD-1 / PD-L1 inhibitor combined with chemotherapy was 3-21 cycles, with an average of 6.23±2.82 cycles. According to the iRECIST standard, 119 patients were evaluated as iCR and iPR, and the treatment effective rate was 51.5%.

[0122] The cycle range of the 97 samples in the internal validation set treated by PD-1 / PD-L1 inhibitor combined with chemotherapy was 3-16 cycles, with an average of 6.26±2.71 cycles. According to the iRECIST standard, 50 patients were evaluated as iCR and iPR, and the treatment effective rate was 51.6%.

[0123] The cycle range of the 40 samples in the external validation set treated by PD-1 / PD-L1 inhibitor combined with chemotherapy was 3-15 cycles, with an average of 6.78±3.10 cycles. According to the iRECIST standard, 18 patients were evaluated as iCR and iPR, and the treatment effective rate was 45.0%.

[0124] 3. CT image acquisition and measurement analysis:

[0125] (1) CT image acquisition:

[0126] All patients underwent chest and abdominal plain scan and dual-phase enhanced scan. The main preparation before scanning included: ① 12 hours of fasting before scanning to empty the gastrointestinal tract; ② 20 minutes of intramuscular injection of 20 mg of anisodamine before scanning to reduce gastrointestinal peristalsis; ③ Within 5-10 minutes before scanning, the patient was asked to drink 800-1200 ml of warm water to ensure that the stomach cavity was fully filled. The CT imaging scheme and scanning parameters are shown in Table 3. The images were reconstructed using standard algorithm, and the slice thickness ranged from 0.625 mm to 5 mm.

[0127]

[0128] (2) Image analysis and measurement

[0129] CT imaging features were measured or assessed independently by two radiologists with 10 (observer A) and 6 (observer B) years of experience in gastrointestinal tumor reading. When there was a disagreement in the qualitative assessment, an expert radiologist was consulted. The quantitative features were based on the average of the two measurements. Before the CT image analysis and measurement, all observers were informed that the study population was gastric cancer patients, but were blinded to other relevant clinical information and treatment.

[0130] In the PACS image diagnosis system, the observers assessed the lesions comprehensively in combination with three-dimensional multiplanar reformation (MPR). The CT imaging features evaluated included: tumor site, tumor CT values in the plain scan, arterial phase, and venous phase, tumor enhancement degree, tumor enhancement pattern, tumor longest diameter, thickest diameter, and clinical TNM stage.

[0131] 1) Tumor site: cardia region, body region, antrum region, ≥2 / 3 of the stomach;

[0132] 2) Tumor CT values in the plain scan, arterial phase, and venous phase;

[0133] 3) Enhancement degree: determined by the difference between the dynamic enhancement CT value and the plain scan CT value, ≤40HU for mild to moderate enhancement, and >40HU for obvious enhancement;

[0134] 4) Tumor enhancement pattern: persistent, regressive, and progressive;

[0135] 5) Tumor longest diameter: the longest diameter line on the largest cross-sectional tumor observed in combination with MPR from multiple angles;

[0136] 6) Tumor thickest diameter: the longest diameter line perpendicular to the circumference of the gastric wall on the axial image;

[0137] 7) Tumor clinical TNM stage: assessed according to the 8th edition of the UICC / AJCC clinical staging system.

[0138] 4. CT image lesion region of interest image (ROI image) delineation

[0139] In this study, the venous phase CT images were selected for radiomics analysis. All the images were uploaded to 3Dslicer software (version 5.0.2, http: / / www.slicer.org). To reduce the influence of different scanning protocols on the robustness of radiomics features, the following pre-processing was performed on all CT images: ① voxel resampling to 1 mm x 1 mm x 1 mm; ② gray value discretization with a group interval of 25; and ③ gray value normalization. Then, an image diagnostician with more than 6 years of experience in gastrointestinal tumor reading used 3D Slicer software (version 5.0.2, http: / / www.slicer.org) to manually segment the ROI on the maximum cross-section of the tumor in the pre-processed venous phase CT images of the stomach tumor region along the lesion edge, and obtain the tumor ROI image Figure 1

[0140] 5. Radiomics feature extraction, selection, and establishment of image scoring model for CT images

[0141] The open-source Python package (Pyradiomics computing language platform) was used to extract features from each ROI image in the training set, a total of 851 radiomics features were extracted, including 107 original features and 744 wavelet transform features. The radiomics features were divided into shape features, first-order statistics features, gray level co-occurrence matrix (GLCM) features, gray level run-length matrix (GLRLM) features, gray level dependence matrix (GLDM) features, neighborhood gray tone difference matrix (NGTDM) features, and gray level size zone matrix (GLSZM) features.

[0142] The intra-class / intra-class consistency coefficient (Inter / intra-class Correlation Coefficient, ICC) was used to evaluate the consistency of radiomics features. Thirty patients were randomly selected, and the ROI was independently drawn by observers A and B to extract radiomics features to evaluate the inter-observer consistency. Observer B repeated the ROI drawing and extracted the radiomics features again after 2 weeks to evaluate the intra-observer consistency.

[0143] ​After feature extraction, the 851 extracted radiomics features were standardized using the Scikit-Learn toolkit in the open-source software python, and then the 851 standardized radiomics features were screened. The specific screening operation was as follows: ① retaining radiomics features with strong robustness (ICC > 0.750); ② secondly, removing highly correlated (Pearson correlation coefficient > 0.95) redundant features by Pearson correlation test; ③ finally, selecting an optimal radiomics feature subset with low collinearity by least absolute shrinkage and selection operator method (LASSO) through 10-fold cross-validation. 439 redundant radiomics features with poor stability were removed by ICC analysis and Pearson correlation analysis, and 412 features were retained for LASSO regression analysis. Finally, 15 optimal features were selected to construct the radiomics feature, as shown in Table 4.

[0144]

[0145] According to the weight coefficient of the 15 screened radiomics features, a lasso regression image score prediction model was established, and the image score prediction model was specifically as follows:

[0146] Radscore = 0.049124367 - 0.108127438 x Original shape Flatness - 0.00194955 x Wavelet LLH glcm Imc1 - 0.011553971 x Wavelet LLH ngtdm Busyness - 0.023343715 x Wavelet LHL glcm Autocorrelation + 0.070103348 x Wavelet LHH firstorder Kurtosis + 0.032363627 x Wavelet HLL glcm MaximumProbability - 0.271672778 x Wavelet HLLglszm GrayLevelNonUniformity - 0.237148842 x Wavelet HLL ngtdm Contrast + 0.205449843 x Wavelet HLH firstorder Kurtosis + 0.262356661 x Wavelet HLHfirstorder Median - 0.073857162 x Wavelet HHL glcm ClusterShade + 0.024585599 x Wavelet HHL glcm InverseVariance - 0.022673688 x Wavelet HHL glszm GrayLevelVariance - 0.062524036 x Wavelet LLL firstorder Maximum + 0.187431418 x Wavelet LLL glszm GrayLevelNonUniformityNormalized. The values of the extracted 15 radiomics features are input into the radiomics score prediction model calculation formula, and the image score can be obtained.

[0147] The image score (denoted as Radscore) of the CT image of the gastric tumor region of each gastric cancer patient in the training set is predicted by using the constructed radiomics score prediction model. The Radscore distribution of each data set is as shown in Figure 2 Figure 2 ​It can be seen that in the training set, the Radscore of the treatment effective group was 0.83±0.37, and the Radscore of the treatment ineffective group was 0.20±0.51, and the difference between the two groups was statistically significant (P<0.001); in the internal validation set, the Radscore of the treatment effective group was 0.74±0.55, and the Radscore of the treatment ineffective group was 0.01±0.66, and the difference between the two groups was statistically significant (P<0.001); in the external validation set, the Radscore of the treatment effective group was 0.75±0.52, and the Radscore of the treatment ineffective group was 0.09±0.72, and the difference between the two groups was statistically significant (P=0.001).

[0148] The ROC curve was drawn according to the image score predicted by the training set and the efficacy grouping of the immunotherapy combined with chemotherapy corresponding thereto, the area under the ROC curve (AUC) was obtained, and the median of the Radscore of the patients in the training set was used as the best cutoff value (i.e. the preset threshold) for distinguishing the treatment effective group and the treatment ineffective group. The preset threshold of the image score prediction model of the application is 0.54.

[0149] 6, Performance evaluation of the image score prediction model constructed by the application:

[0150] The performance of the image score prediction model was verified in the training set, the internal validation set and the external validation set, respectively.

[0151] The image score of the CT image of the tumor region in the stomach of each gastric cancer patient in the training set was predicted by using the constructed image score prediction model, and the ROC curve (the ROC curve is shown in Figure 3 The median of the Radscore of the patients in the training set was used as the best cutoff value (i.e. the preset threshold) for distinguishing the treatment effective group and the treatment ineffective group of the immunotherapy combined with chemotherapy, and the sensitivity and specificity (as shown in Table 5) were calculated, which were used to evaluate the discrimination performance of the score prediction model.

[0152] The values of the imageomic features of the sample CT images in the internal validation set and the external validation set were substituted into the above-mentioned constructed score prediction model to obtain the image score of each sample, and the ROC curve of the image score prediction model was drawn according to the predicted image score (as shown in Figure 4 ), to verify the value of the image score prediction model of the application for distinguishing the treatment effective group and the treatment ineffective group of the immunotherapy combined with chemotherapy. According to the best cutoff value of the score prediction model, the corresponding sensitivity and specificity (as shown in Table 5) were calculated.

[0153] Table 5 The performance of the image score prediction model in discriminating the effective group and the ineffective group of immunotherapy combined with chemotherapy in the training set, the internal validation set and the external validation set

[0154] Sample set AUC Sensitivity (%) Specificity (%) Accuracy (%) Training set 0.868 (95% CI 0.823-0.913) 78.2 82.1 80.1 Internal validation set 0.816 (95% CI: 0.728-0.904) 70.0 91.5 80.4 External validation set 0.793 (95% CI: 0.650-0.936) 66.7 86.4 77.5

[0155] From Figure 3 , Figure 4 and Table 5, the image score prediction model has good performance in discriminating the efficacy of immunotherapy combined with chemotherapy, and the AUC values in the training set, the internal validation set and the external validation set are 0.868, 0.816 and 0.793 respectively, and the accuracy is as high as 80%.

[0156] In order to further verify the performance of the image score prediction model constructed by the present application, the present application further screens the clinical independent prediction factor capable of predicting the efficacy of immunotherapy combined with chemotherapy for gastric cancer patients from the training samples, and combines the screened clinical independent prediction factor with the image score prediction model constructed by the present application, and adopts five different machine learning models to construct the joint prediction model, the five different machine learning models including Logistic regression, eXtreme Gradient Boosting (XGB), support vector machine (SVM), naive bayesian (NB) and random forest (RF), and five joint prediction models are obtained, and the five joint prediction models are respectively: Logistic joint prediction model, XGB joint prediction model, SVM joint prediction model, RF joint prediction model and NB joint prediction model.

[0157] Among them, the screening method of the clinical independent prediction factor for predicting the efficacy of immunotherapy combined with chemotherapy for gastric cancer patients is: based on the clinical data of the patients in the training set of Table 1, the clinical characteristics with P<0.20 are screened by single factor Logistic regression analysis. Further, the clinical characteristics screened by single factor Logistic regression analysis are subjected to multi-factor Logistic regression analysis to screen the clinical independent prediction factor capable of predicting the efficacy of immunotherapy combined with chemotherapy for gastric cancer patients. The results of single factor and multi-factor Logistic regression analysis are shown in Table 6.

[0158]

[0159]

[0160] From Table 6, the clinical characteristics with P<0.20 screened by single factor Logistic regression analysis were: treatment cycle, tumor site, venous phase CT value, platelet, neutrophil, PLR and CA72-4. The clinical independent predictive factors for predicting the efficacy of immunotherapy combined with chemotherapy in patients with gastric cancer screened by multi-factor Logistic regression analysis were treatment cycle (OR=1.20; 95%CI: 1.07-1.35; P=0.001) and CA72-4 (OR=0.50; 95%CI: 0.28-0.91; P=0.024).

[0161] The performance of the above-mentioned five combined prediction models in predicting the efficacy of immunotherapy combined with chemotherapy in gastric cancer was evaluated by the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. Delong test was used to compare the AUC between different models. According to the AUC value and Delong test result, the best machine learning model was determined and a nomogram was drawn. The integrated discrimination improvement (IDI) and categorical net reclassification improvement (NRI) were used to evaluate the additional value of the combined model compared with the imaging-based model in predicting the efficacy. The calibration curve was drawn to test the consistency between the predicted probability and the true probability of the model, and the Hosmer-Lemeshow goodness-of-fit test was used to evaluate whether there was overfitting. The decision curve analysis (DCA) was used to evaluate the clinical practicability of the model.

[0162] In the training set, the AUC value of the XGB combined prediction model was the highest, which was 0.991 (0.979-1.00). In the internal and external validation sets, the AUC values of the Logistic combined prediction model were the highest, which were 0.831 (95%CI: 0.749-0.913) and 0.826 (95%CI: 0.684-0.967), respectively. In terms of stability, taking the AUC value of the training set as the baseline, the AUC value change rates of the XGB combined prediction model in the internal and external validation sets were 23.3% and 28.7%, respectively; the AUC value change rates of the Logistic combined prediction model in the internal and external validation sets were 6.2% and 6.8%, respectively.

[0163] The results of Delong test showed that in the training set, the AUC value of the XGB joint prediction model was significantly higher than those of the Logistic, SVM, RF and NB joint prediction models (all P<0.05); the RF joint prediction model was second, and the AUC value was significantly higher than those of the Logistic, SVM and NB joint prediction models (all P<0.05), and there was no statistically significant difference in the AUC value between the other three models (all P>0.05); in the internal and external validation sets, there was no statistically significant difference in the AUC value between any two models (all P>0.05) (as shown in Table 7).

[0164] The performance statistics of the five joint prediction models are shown in Table 7.

[0165]

[0166]

[0167] Based on the above results, among the five different joint prediction models, the Logistic joint prediction model had a higher and most stable AUC value in the three data sets.

[0168] Comparison of the performance of the Logistic joint prediction model and the image score prediction model constructed in the application:

[0169] The image score prediction model constructed in the application was further compared with the Logistic joint prediction model. The AUC values of the Logistic joint prediction model in the training set (0.886 vs 0.868), internal validation set (0.831 vs 0.816) and external validation set (0.826 vs 0.793) were all slightly higher than those of the image score prediction model of the application. The results of Delong test showed that there was no statistically significant difference between the two models in the training set (Z=1.872, P=0.061), internal validation set (Z=1.139, P=0.255) and external validation set (Z=0.927, P=0.354).

[0170] The integrated discrimination improvement (IDI) and categorical net reclassification improvement (NRI) were used to evaluate the additional diagnostic value of the image score prediction model of the application compared with the Logistic joint prediction model, and the results are shown in Table 8.

[0171] From Table 8, it can be seen that after adding the treatment cycle and CA72-4 clinical variables to the image score prediction model of the present application, the overall prediction ability in the training set is improved, but the effect of specific reclassification of patients is not significant; in addition, in the internal and external validation sets, the clinical variables do not significantly improve the efficacy prediction performance.

[0172]

[0173] The calibration curve and the decision curve analysis (DCA) are drawn to evaluate the fitting degree and clinical benefit of the model, respectively.

[0174] The calibration curve (as shown in Figure 5 ) shows that the prediction results of the image score prediction model of the present application and the Logistic joint prediction model for evaluating the efficacy of immunotherapy combined with chemotherapy in gastric cancer patients are basically consistent with the true results, and have good consistency. The Hosmer-Lemeshow test shows that the difference between the prediction results and the true results of the model is not statistically significant (all P>0.05), and the model fitting is good.

[0175] The DCA decision curve (as shown in Figure 6 ) shows that when the treatment efficacy of gastric cancer is predicted based on the image score prediction model of the present application and the Logistic joint prediction model and the corresponding clinical intervention is performed, the patients can obtain clinical net benefit in a wide range of threshold probability. In the training set, the threshold probability of the two models can reach 0-100%; in the internal validation set, the threshold probability of clinical benefit of the two models is 35%-100%; in the external validation set, the threshold probability of clinical benefit of the two models is 25%-100%.

[0176] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting the efficacy of immunotherapy combined with chemotherapy for gastric cancer based on CT imaging, characterized in that, Includes the following steps: S1. Process medical images of the gastric tumor region of a gastric cancer patient to obtain a region of interest image of the lesion in the medical image of the tumor region; wherein, the medical image is a CT image; S2. Extract the radiomics features from the region of interest image; S3. Input the values ​​of the radiomics features extracted in step S2 into the image scoring prediction model to calculate the image score of the region of interest. S4. Compare the image score obtained in step S3 with a preset threshold, and predict the efficacy of immunotherapy combined with chemotherapy for the gastric cancer patient based on the relationship between the image score and the preset threshold.

2. The method according to claim 1, characterized in that, In step S2, the radiomics features are as follows: Original shape flatness Wavelet LLH glcm Imc1、 Wavelet LLH ngtdm Busyness、 Wavelet LHL glcmAutocorrelation, Wavelet LHH firstorder Kurtosis、 Wavelet HLL glcm MaximumProbability, Wavelet HLL glszm GrayLevelNonUniformity, Wavelet HLL ngtdm Contrast、 Wavelet HLH firstorder Kurtosis、 Wavelet HLH firstorder Median、 Wavelet HHL glcm ClusterShade, Wavelet HHL glcm InverseVariance, Wavelet HHL glszm GrayLevelVariance, Wavelet LLL firstorder Maximum、 Wavelet LLL glszm GrayLevelNonUniformityNormalized.

3. The method according to claim 2, characterized in that, In step S3, the image rating prediction model uses the following formula to calculate the image rating: Radscore=0.049124367-0.108127438×Original shape Flatness-0.00194955×Wavelet LLH glcm Imc1-0.011553971×Wavelet LLH ngtdm Busyness-0.023343715×Wavelet LHL glcm Autocorrelation+0.070103348×Wavelet LHH firstorder Kurtosis+0.032363627×Wavelet HLL glcm MaximumProbability-0.271672778×Wavelet HLLglszm GrayLevelNonUniformity-0.237148842×Wavelet HLL ngtdm Contrast+0.205449843×Wavelet HLH firstorder Kurtosis+0.262356661×Wavelet HLH firstorder Median-0.073857162×Wavelet HHL glcm ClusterShade+0.024585599×Wavelet HHL glcm InverseVariance-0.022673688×Wavelet HHL glszmGrayLevelVariance-0.062524036×Wavelet LLL firstorder Maximum+0.187431418×Wavelet LLL glszm GrayLevelNonUniformityNormalized.

4. The method according to claim 1, characterized in that, In step S4, when the image score is greater than a preset threshold, the predicted efficacy of immunotherapy combined with chemotherapy for gastric cancer patients is effective; when the image score is less than or equal to the preset threshold, the predicted efficacy of immunotherapy combined with chemotherapy for gastric cancer patients is ineffective.

5. The method according to claim 1, characterized in that, The immunotherapy is performed using PD-1 inhibitors and / or PD-L1 inhibitors.

6. The method according to claim 1, characterized in that, The gastric cancer is unresectable gastric cancer; the CT image is a venous phase CT image.

7. A predictive system for the efficacy of immunotherapy combined with chemotherapy for gastric cancer based on CT imaging, characterized in that, The prediction system includes an image input module, an image processing module, a feature extraction module, a scoring prediction module, and an image classification module. The image input module is used to input medical images of the gastric tumor region from a gastric cancer patient. The image processing module processes the input medical images of the gastric tumor region to obtain a region of interest (ROI) image of the lesion. The feature extraction module extracts radiomics features from the ROI image. The scoring prediction module includes a scoring prediction model to calculate the image score of the ROI image based on the values ​​of the radiomics features extracted by the feature extraction module. The prediction and result output module performs qualitative analysis on the image score, predicts the efficacy of immunotherapy combined with chemotherapy in gastric cancer patients, and outputs the prediction results.

8. The prediction system according to claim 7, characterized in that, The radiomics features are as follows: Original shape flatness Wavelet LLH glcm Imc1、 Wavelet LLH ngtdm Busyness、 Wavelet LHL glcmAutocorrelation, Wavelet LHH firstorder Kurtosis、 Wavelet HLL glcm MaximumProbability, Wavelet HLL glszm GrayLevelNonUniformity, Wavelet HLL ngtdm Contrast、 Wavelet HLH firstorder Kurtosis、 Wavelet HLH firstorder Median、 Wavelet HHL glcm ClusterShade, Wavelet HHL glcm InverseVariance, Wavelet HHL glszm GrayLevelVariance, Wavelet LLL firstorder Maximum、 Wavelet LLL glszm GrayLevelNonUniformityNormalized.

9. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, the memory storing a processor-readable application program, the processor being configured to execute the application program to implement the method for predicting the efficacy of gastric cancer immunotherapy combined with chemotherapy based on CT images, as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is loaded by a processor to execute the method for predicting the efficacy of gastric cancer immunotherapy combined with chemotherapy based on CT images, as described in any one of claims 1-6.

Images